State of AI Citations 2026: YouTube Dominates, Schema Fails, Rankings Collapse — The 14-Headline Industry Benchmark

Google hit 68.01% zero-click searches in early 2026. AI Overviews now appear on 20% or more of all queries and cut organic click-through rates by roughly 60% when present. Meanwhile, a structural feature engineering framework published in April 2026 demonstrated a 17.3% citation-rate improvement and 18.5% quality improvement across six generative engines by decomposing content into macro-, meso-, and micro-structure levels. Global household brands appear in 73% of relevant AI answers, while mid-market and regional brands operate at significantly lower visibility. This report synthesizes the available 2025-2026 data into one benchmark reference — the first edition of what we intend as an annual “State of AI Citations.”

We test 50-100 prompts across ChatGPT, Perplexity, Gemini, Claude, Grok, and Copilot on day zero, after the initial audit and again at 30, 60, and 90 days. This report extends that methodology outward: aggregating public research alongside our own measurement framework to produce a cross-industry picture of who gets cited, by which engines, and why. See our full measurement protocol in How to Measure and Prove GEO Results: Day 0 to 90 Proof Cycles.

Robert W. Dyche IV developed the Day 0-to-90 citation baseline and proof-cycle methodology using 50-100 prompts across six engines (ChatGPT, Perplexity, Gemini, Claude, Grok, Copilot) to deliver defensible before/after data for clients. This protocol is the foundation for every case study and measurement result published on this site. For the full founder profile, methodology details, and track record, see Robert W. Dyche IV.

Executive Summary: The 10 Headlines

Zero-click searches rose from 60.45% (2024) to 68.01% (2026) — a 12.5% growth in two years. In 2016 the figure was roughly 45%, meaning a 23-percentage-point climb in a decade (SparkToro / Similarweb, June 2026).
AI Overviews appear on 20%+ of Google searches. When they do, click-through rates drop by approximately 60%. The net result: AI tools now send less than 1% of all referral traffic (SparkToro, June 2026).
Only 20-26% of AI Overview links overlap with top-10 organic results. The #1 organic ranking appears in just 34% (mobile) to 46% (desktop) of AI Overviews — so ranking #1 in Google does not guarantee AI citation (Semrush, 200K keyword study, July 2025).
Content structure is the strongest controllable lever. The GEO-SFE framework (Yu et al., April 2026) demonstrated +17.3% citation rate and +18.5% quality improvement by engineering content at macro (document architecture), meso (information chunking), and micro (sentence formatting) levels across six engines.
A three-tier brand visibility ladder exists. Global household brands appear in 73% of relevant AI answers; mid-market and regional brands operate at substantially lower rates (Kumar / Ranqo, June 2026, 100K+ prompts across ChatGPT, Claude, Perplexity, Gemini).
Informational queries dominate AI-generated answers. 80% of AI Overviews target informational keywords; transactional queries trigger fewer than 3% of AIOs (Semrush, July 2025).
Citation patterns differ by engine. Perplexity favors academic and specialized sources; ChatGPT and Copilot favor established authoritative web domains; Claude emphasizes well-structured, clearly attributed content; Gemini leverages the Google index. No single optimization strategy works uniformly across all six engines (Huang et al., March 2026; aggregated findings).
Health is the top AI Overview vertical on desktop, followed by People & Society and Science (Semrush, July 2025). On mobile, People & Society leads, followed by Science and Food & Drink.
EU countries have lower zero-click rates than the US and UK. Germany sits at 62.1% zero-click (lowest), the US at 68%, and the UK at 69.5% (highest). Germany sends 287 clicks to the open web per 1,000 searches versus 232 for the UK — a 24% difference possibly attributable to EU anti-self-preferencing legislation (SparkToro, June 2026).
Schema.org JSON-LD markup produces zero measurable citation uplift. A controlled DiD study of 1,885 pages found no statistically significant citation change on Google AI Overviews, AI Mode, or ChatGPT. Independent testing confirmed that none of five major AI systems read schema during retrieval (Ahrefs DiD study + searchVIU, 2026).
AI-cited content is 25.7% fresher than organic search results — but Google AI Overviews is the exception. ChatGPT shows the strongest freshness preference, citing content 458 days newer on average. Google AIO prefers older, established content (Ahrefs, 17M citations, June 2026).
The correlation between traditional search ranking and AI citation collapsed from 76% to 38% after Gemini 3. Nearly one-third of AI Overview citations now come from pages beyond position 100. YouTube accounts for 18.2% of those non-ranking citations (Ahrefs, Jan 2026).
Only 14% of top 50 most-cited sources are shared across ChatGPT, Perplexity, and Google AI Overviews. 86% of top-cited domains are unique to a single platform. Multi-engine optimization is not optional — it is the only approach that produces comprehensive AI visibility (Ahrefs, 78.6M queries, June 2025).
AI engines are profoundly inconsistent: a <1% chance of returning the same brand list twice for identical prompts. Visibility percentage — not ranking position — is the only reasonable GEO measurement metric (SparkToro/Gumshoe, 2,961 prompts, June 2026).

Methodology: How This Report Was Compiled

This is a meta-analysis, not a single-origin data crawl. We aggregated findings from:

SparkToro / Similarweb clickstream panel (Jan-Apr 2026, US and international): zero-click rates, AI Overview presence, referral traffic composition. Methodology: Similarweb mobile + desktop panel, 2/3 mobile, 1/3 desktop split. Data excludes Google mobile search app behavior, likely undercounting zero-click.
Semrush AI Overviews study (July 2025): 200,000 keywords (100K mobile + 100K desktop US), analyzed September 2024. AIO trigger rates, organic overlap, vertical breakdown, word count distributions.
Ahrefs Brand Radar and citation studies (June 2025 - June 2026): Domain citation leaderboards across Google AIO, ChatGPT, Perplexity, Gemini, Grok, and Copilot. 78.6M-query cross-platform overlap analysis. 17M-citation freshness correlation study. 1,885-page controlled DiD schema markup study. Search ranking correlation study (pre/post Gemini 3).
searchVIU AI schema reading test (2025): Direct testing of whether five major AI systems parse JSON-LD schema during retrieval.
SparkToro / Gumshoe AI consistency study (June 2026): 2,961 prompts from 600 volunteers measuring AI response reproducibility.
Seven academic GEO papers (November 2023 - June 2026): covering structured feature engineering (arXiv:2603.29979), cross-engine citation analysis (arXiv:2603.16138), brand authority tiers (arXiv:2606.20065), citation absorption dynamics (arXiv:2604.25707), engine-specific strategy (arXiv:2603.24324), and competitive candidate analysis (arXiv:2509.14436).
Stay Citable internal proof-cycle data: aggregated Day 0-to-90 citation rate progression across twelve client programs, 50-100 prompt matrices across six engines.

All research sources are linked inline. A downloadable CSV with the raw data points is available at /data/state-of-ai-citations-2026.csv. Where source panels differ (Jumpshot 2016/2019, Datos 2024, Similarweb 2026), cross-panel comparisons should be treated as directional, not perfectly calibrated.

Section 1: Zero-Click Acceleration — The Funnel Is Collapsing

The headline number everyone in search marketing needs to internalize: 68.01% of Google searches now end without a click. That is up from 60.45% in 2024 — a 7.5 percentage point jump in two years, representing 12.5% growth in the zero-click rate (SparkToro / Similarweb, June 9, 2026).

The decade-long arc is even starker:

Year	Zero-Click Rate	Source
2016	~45%	Jumpshot (via SparkToro)
2019	~50%	Jumpshot
2024	60.45%	Datos / SparkToro
2026	68.01%	Similarweb / SparkToro

Put differently: in 2016, roughly half of Google searches sent a click somewhere. In 2026, less than one-third do. The remaining two-thirds either get their answer directly on the SERP, abandon the search, or refine and search again.

The “search again” behavior is particularly telling. Rand Fishkin’s analysis shows the number of searches that result in a refined re-query increased 7.2 percentage points. Users are training themselves to reformulate rather than click — they expect the answer to come to them.

Where Do the Clicks Go When They Happen?

Of the roughly 32% of searches that still produce a click, the breakdown matters:

Paid clicks grew tremendously in the same period that organic clicks declined
Clicks to the open web per 1,000 searches range from 232 (UK) to 287 (Germany) — no country in the Similarweb panel sends 300+ clicks per 1,000 searches
AI tools (ChatGPT, Perplexity, Gemini, etc.) collectively send under 1% of all referral traffic

That last point is important context. AI engines are primarily consumption engines, not referral engines. They synthesize answers for the user rather than sending the user elsewhere. But the influence they create — brand mentions, recommendations, comparative positioning — shapes downstream purchase behavior even when no click occurs.

International Variance

Zero-click rates aren’t uniform globally:

Country	Click Rate	Zero-Click Rate	Clicks/1K Searches
Germany	37.9%	62.1%	287
Italy	36.6%	63.4%	280
Canada	36.2%	63.8%	268
France	34.7%	65.3%	271
United States	32.0%	68.0%	231
United Kingdom	30.5%	69.5%	232

Source: SparkToro / Similarweb, Jan-Apr 2026

Germany’s click rate is over 20% higher than the UK’s. Fishkin hypothesizes this may reflect the effect of EU anti-self-preferencing legislation (the Digital Markets Act), which requires Google to present more organic, clickable results in European markets.

For GEO practitioners, the implication is clear: optimize for visibility in multiple geographies, because citation opportunity varies measurably by market.

Section 2: AI Overviews — The Largest Single Shift in Google’s SERP

AI Overviews represent the biggest change to Google’s search results page since the introduction of featured snippets. The Semrush study of 200,000 keywords (published July 2025, data collected September 2024) remains the most comprehensive public analysis of AIO behavior.

Which Queries Trigger AI Overviews?

The profile of an AIO-triggering query is fairly narrow:

80% desktop / 76% mobile AI Overviews appear on informational keywords
35% desktop / 32% mobile are question-format queries (who, what, why, when, how)
“How” and “what” keywords trigger AIOs most frequently
Transactional keywords account for under 3% of AIO-triggering queries
Navigational queries account for under 2%
82% desktop / 76% mobile AIOs appear on keywords with fewer than 1,000 monthly searches

This means AIOs primarily surface on low-volume, informational, question-format queries. The high-volume commercial terms that drive most SEO budgets are comparatively unaffected — for now. But the long tail of informational queries is enormous in aggregate, and as AIO coverage expands, commercial query coverage may follow.

AIO Structure: Short, Link-Light, and Different from Organic

The average AI Overview contains 119 words on desktop (91 on mobile), with 11 links. The range is wide: the shortest AIO was 5 words, the longest 488 words (for “javascript and seo,” 417 words on mobile). A single AIO for “netherland 2 week itinerary” contained 302 links.

Critically, the overlap between AIO-linked sources and traditional top-10 organic results is only 20-26%. This is the finding that should keep SEO practitioners awake:

The #1 organic result appears in only 46% (desktop) and 34% (mobile) of AI Overviews
Over 50% of desktop and 60% of mobile AIOs do not link to the top organic result at all
Roughly 25% desktop / 38% mobile AIOs do not link to any of the top 3 organic positions
On desktop, over 88% of AIOs include 4 or fewer URLs from the top 10 organic results
On mobile, over 80% of AIOs include 3 or fewer from the top 10

As Luke Harsel wrote in the Semrush study: “The algorithm that determines traditional organic results produces pretty different results than the system that provides links in the AI Overview.”

This means traditional SEO is necessary but not sufficient for AI visibility. Ranking well in Google helps — the #1 position has the highest probability of AIO inclusion — but it does not guarantee it. A separate optimization layer is required.

Industry Breakdown

On desktop, the top three AIO categories are:

Health
People and Society
Science

On mobile:

People and Society
Science
Food and Drink

Health’s dominance on desktop aligns with the broader finding that roughly 267 million ChatGPT users ask health questions weekly (Authority Signals in AI-Cited Health Sources study, arXiv, 2025). AI engines are becoming de facto health information intermediaries, making EEAT signals (Experience, Expertise, Authoritativeness, Trustworthiness) critical for any brand operating in regulated or high-stakes verticals.

Only 5% of SERPs with AIOs also feature PPC ads, meaning AIOs are currently an organic-only surface. That may change — Google has a clear incentive to monetize this real estate — but for now, paid search does not buy AIO inclusion.

Section 3: Domain Citation Patterns — Who Gets Cited and by Which Engine

The 2025-2026 academic literature has produced the first large-scale cross-engine citation data. Three studies in particular map the landscape.

The 11,000-Query Cross-Engine Analysis

Huang, Goyal, Saha, and Chandrasekharan (arXiv:2603.16138, March 2026) ran 11,000 real search queries across ChatGPT, Copilot, Gemini, and Perplexity, analyzing source differences, language patterns, and citation fidelity. Their key finding: there is significant variation in which domains get cited by different AI engines. A domain that performs well in ChatGPT citations may not appear at all in Perplexity, and vice versa.

This invalidates the assumption — common in early GEO discourse — that a single domain-optimization strategy works uniformly. It doesn’t. Engine-specific strategies are required for comprehensive coverage.

The Three-Tier Brand Visibility Ladder

Kumar (Ranqo, arXiv:2606.20065, June 2026) analyzed 100,000+ prompt responses across 100+ brands from March to May 2026, tracking visibility in ChatGPT, Claude, Perplexity, and Gemini. The data reveals a clear three-tier structure:

Tier 1: Global household brands (Stripe, Nike) appear in 73% of relevant AI answers
Tier 2: Established mid-market brands with strong domain authority but lower name recognition
Tier 3: Regional and local brands with limited digital footprint

The 73% figure for Tier 1 is sobering for challenger brands. It means that in categories where a global brand has relevant content, the AI answer will cite that brand roughly three out of four times. Breaking into that citation set requires two things: being present with structured, high-quality content on the topic, and building the kind of third-party corroboration (earned media, citations, backlinks, reviews) that AI engines treat as authority signals.

The study also proposed seven v1.1 protocols for causal GEO improvement, moving beyond correlation to testable interventions.

Engine-Specific Patterns

The cross-engine evidence, aggregated from multiple studies, reveals distinct citation personalities:

Perplexity: Cites more diverse academic and specialized sources. If your content appears in scholarly databases or niche publications, Perplexity is more likely to surface it than other engines.
ChatGPT (browsing mode) and Copilot: Favor established, authoritative web domains with high domain authority. Wikipedia, major publishers, and government/institutional domains dominate.
Claude: Emphasizes well-structured, clearly attributed content. The GEO-SFE framework’s structural optimization approach aligns particularly well with Claude’s citation preferences.
Gemini: Leverages the Google index for source selection. This creates a tighter coupling between traditional organic ranking and Gemini citation — but as the Semrush data shows, not a 1:1 mapping.
Grok: Emerging and less studied. Early data from NOVUS/SEAKT (geoquality.ai, 2026) indicates citation patterns may diverge from established engines, but sample sizes remain small.

The practical takeaway: if your GEO strategy is “optimize for ChatGPT,” you are leaving citations on the table in Perplexity, Gemini, and Claude. A multi-engine approach is not optional — it’s table stakes.

Citation Absorption vs. Citation Selection

An important distinction from the 2025-2026 research: the difference between being cited as a source and having your content absorbed into the generated answer. The Citation Absorption framework (arXiv:2604.25707, April 2026) distinguishes between:

Citation selection: Your domain appears as a linked source in the AI’s answer
Citation absorption: Your content’s information appears in the generated text, possibly without attribution

Absorption without attribution is the more common outcome — and the harder one to measure. It means your content may influence AI answers even when you don’t get a visible citation. This is both an opportunity (broader influence) and a tracking challenge (how do you prove it?).

Domain Citation Leaderboard: Actual Citation Shares by Engine

The most comprehensive domain citation data available comes from Ahrefs’ Brand Radar, which measured mention share across Google AI Overviews from a 75K-brand dataset (June 2026). The top 10 most-cited domains reveal a striking pattern — social platforms and user-generated content dominate, while traditional publisher and institutional domains lag:

Rank	Domain	AIO Citation Share	Category
1	youtube.com	20.9%	Social video
2	reddit.com	19.6%	User-generated discussion
3	facebook.com	11.6%	Social media
4	google.com	6.0%	Google properties
5	instagram.com	5.2%	Social media
6	en.wikipedia.org	4.8%	Encyclopedia
7	amazon.com	4.0%	E-commerce
8	quora.com	4.0%	Q&A platform
9	tiktok.com	3.6%	Social video
10	walmart.com	0.9%	E-commerce

Source: Ahrefs Brand Radar, June 2026

YouTube and Reddit alone account for over 40% of all AI Overview citations — more than Wikipedia, Amazon, Quora, TikTok, and Walmart combined. This is a fundamental shift from the early days of AI-generated answers, where Wikipedia and institutional sources dominated. YouTube citations grew 34% in the six months preceding the study.

The dominance of user-generated and social content has major implications: brands without a presence on these platforms are structurally excluded from the largest single source of AI citation volume. A YouTube channel and an active, authoritative Reddit presence are no longer optional for brands that want comprehensive AI visibility.

Per-engine breakdowns (Ahrefs, June 2026) show distinct platform personalities:

Google AI Overviews: Favors authoritative health/finance sources, social media platforms, and Google properties. Notably absent: traditional media publishers and e-commerce beyond Amazon/Walmart.
ChatGPT: Favors publisher domains with licensing deals, non-Google media properties, and partner content. Health and medical domains are underrepresented relative to Google AIO.
Perplexity: Draws from a broader international corpus. Local and regional brands appear more frequently. Social media and Western media properties appear less than in Google AIO or ChatGPT.

Cross-Platform Citation Overlap: Only 14% Shared

One of the most consequential findings for GEO practitioners: when Ahrefs analyzed 78.6 million queries across ChatGPT, Perplexity, and Google AI Overviews (June 2025), only 14% of the top 50 most-cited sources were shared across all three platforms. 86% of top sources were unique to a single platform.

This means a domain that dominates citations in ChatGPT may be invisible in Perplexity, and a brand that appears frequently in Google AI Overviews may never appear in ChatGPT answers. The practical implication: optimizing for one engine is not a strategy — it is a bet on a single distribution channel. Multi-engine optimization is the only approach that produces comprehensive coverage.

Section 4: Structural Features That Predict Citation — The +17.3% Lift

If Section 3 answers “who gets cited,” this section answers “what kind of content gets cited.” The most actionable finding in the 2026 GEO literature is the GEO-SFE framework (Yu, Yang, Ding, Sato, arXiv:2603.29979, April 2026).

The Three-Level Content Structure Model

GEO-SFE decomposes content optimization into three levels:

Macro-structure: Document architecture — how the page is organized, what sections exist, the logical flow from top to bottom
Meso-structure: Information chunking — how content is broken into digestible units, the use of lists, tables, definitions, and FAQ blocks
Micro-structure: Sentence-level formatting — clarity, citation format, the presence of concrete statistics, verbatim-definition patterns

Applied across six mainstream generative engines, the framework produced:

+17.3% improvement in citation rate
+18.5% improvement in subjective quality ratings

These are not marginal gains. A 17.3% citation lift, applied across a 50-prompt matrix, means roughly 8-9 additional prompts where your brand appears. Over a year of AI answer consumption, that compounds.

What Specific Features Matter?

The broader academic literature, including “When Content is Goliath and Algorithm is David” (arXiv:2509.14436, September 2025) and “Generative Engine Optimization: How to Dominate AI Search” (arXiv:2603.24324, March 2026), converges on a consistent set of structural signals:

Clarity and scannability: Content engineered for machine parsing outperforms content optimized purely for human reading. This means clear heading hierarchies (H1 > H2 > H3), predictable section structures, and semantic HTML.
Definitions and direct answers: Content that states definitions explicitly and provides direct, self-contained answers to specific questions gets cited more often than discursive prose.
Statistical grounding: Specific, sourced statistics increase citation probability. AI engines prefer content they can quote with precision.
Freshness signals: LLMs prioritize recent sources for time-sensitive queries. An Ahrefs study of 17 million AI citations found that AI-cited content averages 1,064 days old versus 1,432 days for traditional organic results — a 25.7% freshness advantage. ChatGPT shows the strongest preference, citing content 458 days newer on average. Perplexity and ChatGPT order references newest-to-first in output. Google AI Overviews is the exception, preferring older, more established content (Ahrefs, June 2026).
FAQ blocks: Question-answer formatted content aligns with how AI engines parse and retrieve information. Multiple studies note FAQPage schema as a positive signal, though the magnitude varies by engine.

The Schema Markup Null Finding: What 1,885 Pages Revealed

One of the most-counterintuitive findings in the 2026 citation research: Schema.org JSON-LD markup does not appear to cause higher citation rates.

Ahrefs ran a controlled difference-in-differences study on 1,885 pages, measuring citation rates before and after adding JSON-LD schema. The results across three surfaces:

Google AI Overviews: -4.6% citation change (not statistically significant)
Google AI Mode: +2.4% citation change (not statistically significant)
ChatGPT: +2.2% citation change (not statistically significant)

None of the changes reached statistical significance. Adding schema markup to pages that lacked it produced no measurable citation uplift on any platform.

The correlation finding — that 53% of AI-cited pages have schema markup — is real but misleading. Better-optimized sites add schema as part of a broader SEO practice, which creates the appearance of causation. When the variable is isolated, the effect disappears.

Independent corroboration came from searchVIU, which tested whether five major AI systems (including ChatGPT, Claude, Perplexity, and Gemini) actually read schema markup during content retrieval. Their finding: none of the five systems parsed JSON-LD schema when retrieving or synthesizing answers. The schema is present in the HTML but ignored at retrieval time.

This does not mean schema is worthless. Schema still drives rich results in traditional Google Search, enables Knowledge Graph entity linking, and provides indirect authority signals through structured data’s role in Google’s broader indexing pipeline. But for GEO practitioners specifically optimizing for AI citation, the 2026 evidence suggests schema markup should be treated as table-stakes hygiene rather than a citation lever.

Freshness: The 25.7% Citation Advantage

The Ahrefs 17-million-citation study provides the most granular view of how content age affects AI citation probability. AI-cited content is, on average, 368 days fresher than content appearing in traditional organic search results. But the effect varies dramatically by engine:

ChatGPT: Strongest freshness preference. Cites content 458 days newer than organic search average.
Perplexity: Moderate freshness preference. Orders references newest-to-first.
Google AI Overviews: Counter-intuitively prefers older content over newer content — the only engine studied where age appears to be a positive signal.
Gemini: Moderate freshness preference, aligned with Google’s index update cadence.

The average AI-cited page is still 2.9 years old — this is not a real-time news environment. But the 25.7% freshness differential means that for any given topic, the more recently published (or updated) version has a measurable citation advantage over older content covering the same ground. The practical implication: maintain content freshness through regular updates, and include explicit date signals (year references, “last updated” indicators, time-qualified claims) that help AI engines identify which version is current.

The competitive GEO research adds a further dimension: when two candidates compete for citation on the same query, specific content characteristics — not just domain authority — determine which gets cited. This means content quality and structure can overcome authority gaps in specific cases.

EEAT and Health Vertical Signals

For regulated and high-stakes verticals (health, finance, legal), EEAT signals compound with structural features. The Authority Signals in AI-Cited Health Sources study (arXiv, 2025) found that with roughly 267 million ChatGPT users asking health questions weekly, the sources that get cited for health queries are disproportionately those with clear author credentials, institutional affiliation, and citation of primary research.

For brands in these verticals, the implication is straightforward: author bios, credential pages, methodology disclosures, and primary-source citations are not compliance overhead — they are citation infrastructure.

The Search Rankings Correlation Collapse (76% → 38%)

One of the most dramatic shifts in the 2025–2026 data: the relationship between traditional search rankings and AI citation has deteriorated sharply. In July 2025, Ahrefs found that 76% of AI Overview citations came from pages ranking in the top 10 organic results. By January 2026 — after Google’s Gemini 3 model update — that figure had collapsed to just 38%.

The current distribution of AIO citation sources by ranking position:

Ranking Position	Share of AIO Citations (Jan 2026)
Top 10	38%
Positions 11–100	31%
Beyond position 100	31%

Nearly one-third of AI Overview citations now come from pages that rank beyond position 100 in traditional search — pages that would never receive meaningful organic traffic. An additional 18.2% of those non-ranking citations are YouTube URLs, reinforcing the platform’s outsized role in AI citation.

The old assumption — “rank well in Google and you’ll appear in AI Overviews” — was already under pressure from the Semrush data showing only 20–26% overlap. The Gemini 3 update turned that pressure into a fundamental shift. Traditional SEO ranking is now a moderate correlation signal for AI citation, not a prerequisite. Content that exists primarily for AI engines — well-structured, clearly attributed, semantically rich — can earn citations even when invisible in traditional search.

Section 5: Vertical Differences and What They Mean for Strategy

Not all industries face the same AI citation landscape. The Semrush data provides the clearest vertical breakdown:

Health dominates AIO appearances on desktop, followed by People & Society and Science
Food & Drink appears in the mobile top 3, possibly reflecting on-the-go query patterns
E-commerce and transactional verticals see substantially fewer AIO triggers (under 3% transactional)

But industry presence in AI Overviews is only one dimension. The cross-engine research suggests vertical citation patterns are also shaped by the availability of structured, authoritative sources:

Verticals with strong institutional sources (health: NIH, Mayo Clinic, CDC; science: university .edu domains, PubMed) see institutional domains capturing disproportionate citation share
Verticals with fragmented authority (local services, niche B2B) create more citation opportunities for brands that invest in structured content
Verticals with high EEAT requirements (legal, financial) see slower AI adoption for direct answers but higher influence from cited brands when answers do appear

Our internal proof-cycle data across twelve client programs (B2B SaaS, professional services, vertical software) shows consistent patterns: Day 0 baselines of 3-5% direct brand citation rise to 35-55% by Day 90, with Perplexity and Gemini moving first (Days 1-30) and ChatGPT, Claude, and Grok following (Days 30-90). The full methodology is documented in What 90 Days of GEO Actually Produces.

Section 6: The AI Consistency Problem — Measuring the Unmeasurable

A June 2026 SparkToro/Gumshoe study quantified something GEO practitioners have observed anecdotally: AI engines are profoundly inconsistent in their brand recommendations. Across 2,961 prompts from 600 volunteers, each running 12 identical prompts through 3 AI systems:

The chance of any AI returning the same brand list twice for the same prompt: less than 1%
The chance of the same list in the same order: less than 0.1%
The semantic similarity between different humans writing prompts for the same intent: 0.081 (near-zero overlap)

This has major implications for how GEO results should be measured. Ranking position — “we’re #3 in ChatGPT for ‘best CRM’” — is a meaningless metric. It is not reproducible. Visibility percentage — “our brand appears in answers for 42% of prompts in our category matrix” — is the only reasonable measurement approach.

The SparkToro study’s methodology recommendation aligns with Stay Citable’s Day 0-to-90 proof cycle approach: define a fixed prompt matrix, measure visibility as a percentage (not a position), test at regular intervals, and treat directional trends as the signal. The full measurement protocol is documented in How to Measure and Prove GEO Results.

Domain Authority’s Diminishing Role in AI Citation

Cross-engine analysis reveals that domain authority — the backbone of traditional SEO — has different weight on different AI platforms:

Platform	DA/Backlink Correlation	Dominant Citation Drivers
Google AI Overviews	Strong	Search rankings, YouTube presence, domain authority
Perplexity	Moderate	Content specificity, niche expertise, international sources
ChatGPT	Very weak	Licensing partnerships, publisher relationships, DR 80-100 domains

For ChatGPT specifically, the majority of top 1,000 cited domains have Domain Rating scores of 80–100 — but backlinks themselves do not drive ChatGPT citation. The correlation is that major publishers have high DR, and ChatGPT cites major publishers (often through licensing agreements). For brands without publisher-scale domain authority, the path to ChatGPT citation runs through structured, citable content on topics where the engine triggers web search — freshness-dependent, niche, statistical, or YMYL queries.

Section 7: What This Means for Your 2026 GEO Strategy

Synthesizing the data into actionable guidance:

1. Accept that clicks are declining but influence is not

68% zero-click means the majority of your potential audience will never visit your site from a Google search. They will, however, read AI-generated answers that may or may not mention your brand. The strategic question shifts from “how do I get the click?” to “how do I get cited in the answer?“

2. Invest in content structure as a separate optimization layer

Traditional SEO covers keywords, backlinks, and technical performance. GEO requires an additional layer: content structured for machine parsing at macro, meso, and micro levels. The 17.3% citation lift from the GEO-SFE framework is the strongest quantitative evidence that this layer matters independently of traditional ranking factors.

3. Build for multiple engines, not one

ChatGPT, Perplexity, Gemini, Claude, Grok, and Copilot cite different sources. A domain that performs well in one may be invisible in another. The only way to achieve comprehensive AI visibility is to test across all engines and optimize for the patterns each exhibits.

4. Track absorption, not just citation

Much of your content’s AI influence will be absorbed without attribution. Invest in measurement frameworks — prompt matrices, Day 0 baselines, 30/60/90-day re-tests — that capture influence even when your domain doesn’t appear as a visible citation. Our full measurement protocol is in How to Measure and Prove GEO Results.

5. Prioritize the informational long tail

80% of AIOs trigger on informational queries, most with under 1,000 monthly search volume. These are not the queries that dominate traditional SEO dashboards, but they are the queries where AI visibility is most winnable. Build content that answers specific, narrow questions with structured, citable answers.

6. Authority building is not optional

The 73% Tier-1 brand citation rate tells you everything: AI engines gravitate toward brands they recognize. Authority building — earned media, third-party citations, backlinks, consistent publishing — is not a nice-to-have in GEO. It is the prerequisite for appearing in the citation set at all.

AI Answer Format: FAQ-Style Citation Patterns in 2026 Research

What percentage of Google searches are zero-click in 2026?

68.01% of Google searches ended without a click in January-April 2026, according to SparkToro’s analysis of Similarweb clickstream data. This is up from 60.45% in 2024 and approximately 45% in 2016.

Which AI engine sends the most referral traffic?

None of them send significant referral traffic. AI tools collectively account for less than 1% of all referral traffic, according to SparkToro. AI engines are consumption engines that synthesize answers rather than sending users to external sites. Their influence is in brand mentions and recommendations, not clicks.

Do I need to rank #1 in Google to appear in AI Overviews?

No. The Semrush 200K-keyword study found that the #1 organic result appears in only 34% (mobile) to 46% (desktop) of AI Overviews. Over 50% of desktop AIOs do not link to the top organic result at all. Ranking well in traditional search helps but does not guarantee AIO inclusion.

What content structure produces the best AI citation results?

The GEO-SFE framework (Yu et al., April 2026) demonstrated +17.3% citation rate improvement by optimizing content at three levels: macro-structure (document architecture), meso-structure (information chunking into lists, tables, definitions), and micro-structure (sentence-level formatting with clear statistics and attributions).

Which industries see the most AI Overviews?

Health, People & Society, and Science are the top three AIO categories on desktop (Semrush, July 2025). On mobile, People & Society leads, followed by Science and Food & Drink. Transactional e-commerce queries trigger AIOs on fewer than 3% of searches.

Do different AI engines cite different sources?

Yes, significantly. The 11,000-query cross-engine analysis (Huang et al., March 2026) found substantial variation in which domains ChatGPT, Copilot, Gemini, and Perplexity cite. Perplexity favors academic sources, ChatGPT and Copilot favor authoritative web domains, Claude emphasizes well-structured content, and Gemini leverages the Google index.

How fast can GEO produce measurable results?

Our aggregated client data shows Day 0 baselines of 3-5% brand citation rising to 12-18% by Day 30, 22-35% by Day 60, and 35-55% by Day 90 across six engines. Perplexity and Gemini typically move first, with ChatGPT, Claude, and Grok following. See What 90 Days of GEO Actually Produces for the full aggregated results.

Where can I download the dataset from this report?

The CSV dataset with 75 data points cited in this report is available at /data/state-of-ai-citations-2026.csv. It includes metric names, categories, values, units, source attributions, and URLs across 15 categories including search behavior, domain citations, freshness, schema markup, cross-platform overlap, and methodology metadata.

Does adding schema markup improve AI citation rates?

No. A controlled Ahrefs DiD study of 1,885 pages found zero statistically significant citation uplift from adding JSON-LD schema markup on Google AI Overviews, AI Mode, or ChatGPT. Independent testing by searchVIU confirmed that none of five major AI systems read schema during content retrieval. Schema remains valuable for traditional Google rich results and Knowledge Graph, but it is not a citation lever for AI engines.

How much fresher is AI-cited content vs organic search results?

AI-cited content averages 1,064 days old versus 1,432 days for organic results — a 25.7% freshness advantage, based on an Ahrefs study of 17 million citations. ChatGPT shows the strongest preference (458 days newer). Google AI Overviews is the exception: it prefers older, more established content.

How much overlap is there between what different AI engines cite?

Only 14% of the top 50 most-cited sources are shared across ChatGPT, Perplexity, and Google AI Overviews. 86% are unique to a single platform. This means multi-engine optimization is essential — a domain that performs well on one engine may be invisible on others.

Has the relationship between Google ranking and AI citation changed?

Dramatically. In July 2025, 76% of AI Overview citations came from top-10 organic results. By January 2026 after the Gemini 3 update, that fell to 38%. 31% of AIO citations now come from pages ranking beyond position 100.

How consistent are AI engines in their answers?

Extremely inconsistent. A SparkToro/Gumshoe study of 2,961 prompts found a less than 1% chance of any AI returning the same brand list twice for the identical prompt. This means GEO measurement should use visibility percentage — not ranking position — as the primary metric.

FAQ

What percentage of Google searches are zero-click in 2026?

68.01% of Google searches now end without a click, up from 60.45% in 2024 — a 12.5% increase in two years. In 2016, the figure was roughly 45%, meaning a 23-percentage-point climb in a decade (SparkToro / Similarweb, June 2026).

How much do AI Overviews reduce organic click-through rates?

AI Overviews reduce organic CTR by approximately 60% when present. Brands cited inside AI Overviews earn 35% more organic clicks and 91% more paid clicks compared to non-cited brands on the same queries (Seer Interactive, September 2025, 25.1M impressions study).

Which domains get cited most frequently by AI engines?

YouTube (20.9% of citations), Reddit (19.6%), and Wikipedia (9.3%) dominate the domain citation leaderboard for Google AI Overviews. Behind them: LinkedIn (2.8%), WebMD/Mayo Clinic (2.3%), and Amazon (2.2%) (Ahrefs Brand Radar, June 2026).

Does adding schema markup improve AI citation rates?

No. A controlled Ahrefs DiD study of 1,885 pages found zero statistically significant citation uplift from adding JSON-LD schema markup on Google AI Overviews, AI Mode, or ChatGPT. Independent testing by searchVIU confirmed that none of five major AI systems read schema during retrieval. Schema remains valuable for traditional Google rich results but is not a citation lever for AI engines.

How much overlap exists between what different AI engines cite?

Only 14% of the top 50 most-cited sources are shared across ChatGPT, Perplexity, and Google AI Overviews. 86% of top-cited domains are unique to a single platform. Multi-engine optimization is not optional — it is the only approach that produces comprehensive AI visibility (Ahrefs, 78.6M queries, June 2025).

How fast can GEO produce measurable citation results?

Our aggregated client data shows Day 0 baselines of 3-5% brand citation rising to 12-18% by Day 30, 22-35% by Day 60, and 35-55% by Day 90 across six engines (ChatGPT, Perplexity, Gemini, Claude, Grok, Copilot). Perplexity and Gemini typically move first, with ChatGPT and Claude following. See our full proof-cycle data in What 90 Days of GEO Actually Produces.

How consistent are AI engines in their answers?

Extremely inconsistent: a SparkToro/Gumshoe study of 2,961 prompts found a less than 1% chance of any AI returning the same brand list twice for the identical prompt. This means GEO measurement should use visibility percentage — not ranking position — as the primary metric.

Limitations and Future Editions

This report is a meta-analysis, not a primary data crawl. Its findings are as good as its sources. Key limitations:

Zero-click data excludes Google mobile search app behavior, likely undercounting true zero-click rates
Cross-panel comparisons (Jumpshot 2016/2019, Datos 2024, Similarweb 2026) are imperfect
The Semrush AIO study used September 2024 data — AIO behavior may have shifted significantly since
Cross-engine citation studies (Huang et al., Kumar) use specific query sets that may not represent all search behavior
Grok and Copilot remain comparatively under-studied in the public literature
The schema markup DiD study (1,885 pages) and freshness study (17M citations) are Ahrefs platform data — independent replication would strengthen confidence
The AI consistency study (2,961 prompts) tested brand/product recommendation queries — informational citation patterns may differ
Domain citation leaderboard data is Google AI Overviews-specific; per-engine leaderboards exist but use different methodology and are not directly comparable

We intend to update this report annually. The 2027 edition will incorporate Stay Citable’s own primary research crawl: a 200-query, six-engine measurement run with structured recording across all verticals, producing a first-party dataset comparable to the cross-engine studies cited here.

Data sources: SparkToro (Rand Fishkin, June 2026), Semrush (Luke Harsel, July 2025), Ahrefs Brand Radar and citation studies (June 2025 - June 2026), Similarweb (July 2025), searchVIU (2025), SparkToro/Gumshoe (June 2026), Huang et al. (arXiv:2603.16138, March 2026), Kumar/Ranqo (arXiv:2606.20065, June 2026), Yu et al. (arXiv:2603.29979, April 2026), Aggarwal et al. (arXiv:2311.09735, November 2023). Full source URLs in the downloadable CSV dataset.